Dual-Corpus RAG for Women's Hormonal Health: Component-Level Ablation Studies and Statistical Evaluation
May 2026
CSCI E-222 Foundations of Large Language Models, Harvard Extension School
This project built and evaluated a dual-corpus RAG system for women's hormonal health question answering. The system paired PubMedQA clinical evidence with MedQuAD patient-facing explanations, compared three retrieval modes, ran three ablation studies, and used bootstrap confidence intervals across 17 experimental conditions.
For an annotated code walkthrough, see the Research page.